[R] Generate a matrix Q satisfying t(Q)%*%Q=Z and XQ=W

Wed Jul 7 21:28:45 CEST 2004

Thanks but it does not solve my problem because I would like to generate Q 
(X2) for a given X(X1). X is fixed. But It seems hard to do it and perhaps 
It would be easier to change my approach.

Thanks again for your help

At 14:55 07/07/2004, Spencer Graves wrote:
>      How about generating matrices (X1|X2), dim(X1) = c(n, k1), dim(X2) = 
> c(n, k2), with mean 0 and covariance matrix as follows:
>(S11 | S12)
>(S21 | S22),
>
>with S12 = W, S22 = Z and S11 = whatever you want?  With X = t(X1), and Q 
>= X2, we have E(XQ) = W and E(Q'Q) = Z.
>      This can be done using rmvnorm in package mvtnorm.
>      hope this helps.  spencer graves
>
>Stephane DRAY wrote:
>
>>thanks,
>>I want to create matrices for simulation purpose (in order to evaluate 
>>the efficiency of different methods on this simulated data set). So, I 
>>want to create a data matrix Q (m individuals and r variables). I want to 
>>specify the variance-covariance structure for this matrix (t(Q)%*%Q=Z ) 
>>but I want also to create another constraint due to another matrix of 
>>data. I want that the covariance of Q and X are equal to those given in W 
>>(XQ=W).
>>The example I gave is just to illustrate my problem and perhaps it has no 
>>solution (I cannot see it because I have no idea how to construct Q such 
>>as Q=Q1=Q2)
>>
>>
>>
>>
>>At 00:00 07/07/2004, Spencer Graves wrote:
>>
>>>      Is a solution even possible for the matrices in your example?
>>>I've tried a few things that have suggested that a solution may not be 
>>>possible.
>>>      What can you tell us of the problem that you've translated into 
>>> this?  I see a minimization problem subject to constraints, but I'm not 
>>> certain which are the constraints and what is the objective function.
>>>For example, are you trying to find Q to minimize sum((Z-X'X)^2) subject 
>>>to XQ=W or do you want to minimize sum((XQ-W)^2) subject to Q'Q=Z or 
>>>something else?
>>>      If it were my problem, I think I would work for a while with the 
>>> singular value decompositions of X, W and Z, and see if that would lead 
>>> me to more information about Q, including conditions under which a 
>>> solution existed, expressions for Q when multiple solutions existed, 
>>> and a solution minimizing your chosen objective function when solutions 
>>> do not exist.  (A google search produced many hits for "singular value 
>>> decomposition", implemented as "svd" in R.)
>>>      hope this helps.  spencer graves
>>>
>>>Stephane DRAY wrote:
>>>
>>>>Hello,
>>>>I have a question that is not directly related to R ... but I try to do 
>>>>it in R ;-) :
>>>>
>>>>I would like to generate a matrix Q satisfying (for a given Z, X and W) 
>>>>the two following conditions:
>>>>
>>>>t(Q)%*%Q=Z  (1)
>>>>XQ=W (2)
>>>>
>>>>where:
>>>>Q is m rows and r columns
>>>>X is p rows and m columns
>>>>D is p rows and r columns
>>>>C is r rows and r columns
>>>>with m>p,r
>>>>
>>>>
>>>>e.g:
>>>>m=6,
>>>>p=2
>>>>r=3
>>>>
>>>>Z=matrix(c(1,.2,.5,.2,1,.45,.5,.45,1),3,3)
>>>>X=matrix(c(.1,.3,.5,.6,.2,.1,.8,1,.4,.2,.2,.9),2,6)
>>>>W=matrix(c(0,.8,.4,.6,.2,0),2,3)
>>>>
>>>>#Create a matrix satisfying (1) is easy:
>>>>
>>>>A=matrix(runif(18),6,3)
>>>>Q1=svd(A)$u%*%chol(Z)
>>>>
>>>>
>>>>#For the second condition (2), a solution is given by
>>>>
>>>>Q2=A%*%ginv(X%*%A)%*%W
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>I do not know how to create a matrix Q that satisfies the two 
>>>>conditions.  I have try to construct an iterative procedure without 
>>>>success (no convergence):
>>>>
>>>>eps=10
>>>>i=0
>>>>while(eps>.5)
>>>>{
>>>>Q1=svd(Q2)$u%*%chol(Z)
>>>>Q2=Q1%*%ginv(X%*%Q1)%*%W
>>>>eps=sum(abs(Q1-Q2))
>>>>cat(i,":",eps,"\n")
>>>>i=i+1
>>>>}
>>>>
>>>>Perhaps someone could have any idea to solve the problem, or a 
>>>>reference on this kind of question or the email of another list where I 
>>>>should ask this question.
>>>>
>>>>Thanks in advance,
>>>>
>>>>Sincerely.
>>>>
>>>>Stéphane DRAY
>>>>-------------------------------------------------------------------------------------------------- 
>>>>
>>>>Département des Sciences Biologiques
>>>>Université de Montréal, C.P. 6128, succursale centre-ville
>>>>Montréal, Québec H3C 3J7, Canada
>>>>
>>>>Tel : 514 343 6111 poste 1233
>>>>E-mail : stephane.dray at umontreal.ca
>>>>-------------------------------------------------------------------------------------------------- 
>>>>
>>>>Web
>>>>http://www.steph280.freesurf.fr/
>>>>
>>>>______________________________________________
>>>>R-help at stat.math.ethz.ch mailing list
>>>>https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>>>>PLEASE do read the posting guide! 
>>>>http://www.R-project.org/posting-guide.html
>>>
>>
>>Stéphane DRAY
>>-------------------------------------------------------------------------------------------------- 
>>
>>Département des Sciences Biologiques
>>Université de Montréal, C.P. 6128, succursale centre-ville
>>Montréal, Québec H3C 3J7, Canada
>>
>>Tel : 514 343 6111 poste 1233
>>E-mail : stephane.dray at umontreal.ca
>>-------------------------------------------------------------------------------------------------- 
>>
>>Web
>>http://www.steph280.freesurf.fr/
>>-------------------------------------------------------------------------------------------------- 
>>

Stéphane DRAY
-------------------------------------------------------------------------------------------------- 

Département des Sciences Biologiques
Université de Montréal, C.P. 6128, succursale centre-ville
Montréal, Québec H3C 3J7, Canada

Tel : 514 343 6111 poste 1233
E-mail : stephane.dray at umontreal.ca
-------------------------------------------------------------------------------------------------- 

Web                                          http://www.steph280.freesurf.fr/